the old code worked fine for a long time and was not affected by
the bug the new code fixes and the new is not widely tested yet.
This can be reverted once the code received more testing in
master
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
* commit '517ce1d09b5e6b72afc2ef9490b5f8ca42fa6a65':
lavu: fix memory leaks by using a mutex instead of atomics
Conflicts:
libavutil/buffer.c
The atomics code is left in place as a fallback for synchronization in the
absence of p/w32 threads. Our ABI did not requires applications to
only use threads (and matching ones) to what libavutil was build with
Our code also was not affected by the leak this change fixes, though
no question the atomics based implementation is not pretty at all.
First and foremost the code must work, being pretty comes after that.
If this causes problems, for example when libavutil is used by multiple
applications each using a different kind of threading system then the
default possibly has to be changed to the uglier atomics.
See: cea3a63ba3
Merged-by: Michael Niedermayer <michaelni@gmx.at>
The buffer pool has to atomically add and remove entries from the linked
list of available buffers. This was done by removing the entire list
with a CAS operation, working on it, and then setting it back again
(using a retry-loop in case another thread was doing the same thing).
This could effectively cause memory leaks: while a thread was working on
the buffer list, other threads would allocate new buffers, increasing
the pool's total size. There was no real leak, but since these extra
buffers were not needed, but not free'd either (except when the buffer
pool was destroyed), this had the same effects as a real leak. For some
reason, growth was exponential, and could easily kill the process due
to OOM in real-world uses.
Fix this by using a mutex to protect the list operations. The fancy
way atomics remove the whole list to work on it is not needed anymore,
which also avoids the situation which was causing the leak.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
(cherry picked from commit fbd6c97f9c)
Signed-off-by: Anton Khirnov <anton@khirnov.net>
Also add no-op fallbacks when threading is disabled.
This helps keeping the code clean if Libav is compiled for targets
without threading. Since we assume that no threads of any kind are used
in such configurations, doing nothing is ok by definition.
Based on a patch by wm4 <nfxjfg@googlemail.com>.
(cherry picked from commit 2443e522f0)
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The emulation is unused and causes compilation trouble on systems
where fminf() is defined in <math.h> but missing from libm.
This should fix compilation on Debian powerpcspe.
(cherry picked from commit 4436a8f44d)
Signed-off-by: Philip DeCamp <decamp@mit.edu>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
(cherry picked from commit 857fc0a71f)
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This avoids several failures on fate.ffmpeg.org, and thus makes real
bugs easier to spot
Reviewed-by: James Darnley <james.darnley@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Add the feature test macro which is required for building with the
musl toolchain.
The feature test macro _XOPEN_SOURCE = 600 provides the XSI-compliant
version of strerror_r().
Signed-off-by: Jörg Krause <jkrause@posteo.de>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Unfortunately this was not explicitly documented and thus
might be risky.
But all uses I could find in FFmpeg and one in VLC had a memleak
in these cases, and I could not find any that relied on the previous
behaviour.
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
~560 → ~500 decicycles
This is following the comments from Michael in
https://ffmpeg.org/pipermail/ffmpeg-devel/2014-August/160599.html
Using 2 registers for accumulator didn't help. On the other hand,
some re-ordering between the movs and psadbw allowed going ~538 to ~500.
The reasoning behind this addition is that various third party
applications are interested in getting some motion information out of a
video "for free" when it is available.
It was considered to export other information as well (such as the intra
information about the block, or the quantization) but the structure
might have ended up into a half full-generic, half full of codec
specific cruft. If more information is necessary, it should either be
added in the "flags" field of the AVMotionVector structure, or in
another side-data.
This commit also includes an example exporting them in a CSV stream.
This allows getting rid of the many, slightly differing, implementations
of basically the same thing.
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
